MiMo-V2.6-Flash

Open Source

Open Weights

Parameters

309B

Context Length

1.05M

Modality

Multimodal

Architecture

License

Release Date

21 Sept 2026

Knowledge Cutoff

API Pricing (per 1M)

Input: $0.14 · Output: $0.28

System Requirements

VRAM requirements for different quantization methods and context sizes

1,024 tokens

605.97 GB VRAM

Consumer

33x RTX 4090

24GB VRAM

Datacenter

9x NVIDIA A100

80GB VRAM

Apple Silicon

7x Apple M3 Max

128GB VRAM

1,048,576 tokens

740.24 GB VRAM

Consumer

41x RTX 4090

24GB VRAM

Datacenter

11x NVIDIA A100

80GB VRAM

Apple Silicon

9x Apple M3 Max

128GB VRAM

Architecture Diagram

Evaluation Benchmarks

Rank

#37

Benchmark	Score	Rank
Web Development WebDev Arena	1636	16
Agent Arena Agent Arena	-0.57	39
Intelligence Index Artificial Analysis	0.38	64
General Text Text Arena	1452	66

Rankings

Overall Rank

#37

Coding Rank

#43

About MiMo-V2.6-Flash

MiMo-V2.6-Flash is an open-source multimodal mixture-of-experts foundation model developed by Xiaomi. With 309B total parameters activating 15B per token, it delivers high-efficiency performance across text, image, audio, and video inputs.

Technical Specifications

Attention

Attention Structure

Attention Heads

Key-Value Heads

Attention Head Dimension

Position Embedding

RoPE Theta

Sliding Window Attention

Sliding Window Size

Sliding Window Ratio

Linear Attention

Linear Attention Ratio

Normalization

Activation Function

Dimensions

Auxiliary Parameters

Hidden Dimension Size

Number of Layers

FFN Intermediate Size (Dense)

Multi-Token Prediction Heads

Tokenizer

Vocabulary Size

Resources

Download Weights

About MiMo V2

MiMo-V2-Flash is a Mixture-of-Experts (MoE) model with hybrid attention architecture designed for high-speed reasoning and agentic workflows. It features Multi-Token Prediction (MTP) to achieve state-of-the-art performance while significantly reducing inference costs. The model is optimized for long-context modeling and efficient inference.